TY - GEN
T1 - Neural network predictive control in a naturally ventilated and fog cooled greenhouse
AU - Fitz-Rodríguez, E.
AU - Kacira, M.
AU - Villarreal-Guerrero, F.
AU - Giacomelli, G. A.
AU - Linker, R.
AU - Kubota, C.
AU - Arbel, A.
PY - 2012/6/1
Y1 - 2012/6/1
N2 - Passive ventilation in greenhouse production systems is predominant worldwide, limiting its usability and profitability to specific regions or for short production cycles. Evaporative fogging systems have increasingly been implemented in Arid and Semi-Arid regions to extend the production cycle during the warmest season, and also to achieve near-optimum environments for year-round production. However, appropriate control strategies for evaporative fogging systems are still lacking or limited despite its reported benefits in terms of environmental uniformity and potential savings in water and energy usage, when compared to fan and pad systems. The present research proposes a neural network predictive control approach for optimizing water and energy usage in a naturally ventilated and fog cooled greenhouse while providing a near-optimum and uniform environment for plant growth. As a first step the dynamic behavior of the greenhouse environment, defined by air temperature and relative humidity, was characterized by means of system identification using a recurrent dynamic network (NARMX). The multi-step ahead prediction capability of NARMX allows for the optimization of the control actions (vent configuration and fogging rate) for its implementation in the NN predictive control scheme. Greenhouse environmental data from a set of experiments consisting of several vent configurations (0/50, 0/100, 50/50, 50/100 and 100/100, percent opening of the side/roof vents) and three fogging rates (17.5, 22.3 and 27.0 g m-2min-1) during several days throughout the year were used in the system identification process. The resulting NN model accurately predicted the dynamic behavior of the greenhouse environment, having coefficients of determination (R2) of 0.99 for each parameter (air temperature and relative humidity). These NN model will be incorporated into the NN predictive control scheme and its feasibility is in a naturally ventilated greenhouse equipped with a variable-rate fogging system is discussed, while achieving a greenhouse environment within defined permissible ranges of air temperature and relative humidity.
AB - Passive ventilation in greenhouse production systems is predominant worldwide, limiting its usability and profitability to specific regions or for short production cycles. Evaporative fogging systems have increasingly been implemented in Arid and Semi-Arid regions to extend the production cycle during the warmest season, and also to achieve near-optimum environments for year-round production. However, appropriate control strategies for evaporative fogging systems are still lacking or limited despite its reported benefits in terms of environmental uniformity and potential savings in water and energy usage, when compared to fan and pad systems. The present research proposes a neural network predictive control approach for optimizing water and energy usage in a naturally ventilated and fog cooled greenhouse while providing a near-optimum and uniform environment for plant growth. As a first step the dynamic behavior of the greenhouse environment, defined by air temperature and relative humidity, was characterized by means of system identification using a recurrent dynamic network (NARMX). The multi-step ahead prediction capability of NARMX allows for the optimization of the control actions (vent configuration and fogging rate) for its implementation in the NN predictive control scheme. Greenhouse environmental data from a set of experiments consisting of several vent configurations (0/50, 0/100, 50/50, 50/100 and 100/100, percent opening of the side/roof vents) and three fogging rates (17.5, 22.3 and 27.0 g m-2min-1) during several days throughout the year were used in the system identification process. The resulting NN model accurately predicted the dynamic behavior of the greenhouse environment, having coefficients of determination (R2) of 0.99 for each parameter (air temperature and relative humidity). These NN model will be incorporated into the NN predictive control scheme and its feasibility is in a naturally ventilated greenhouse equipped with a variable-rate fogging system is discussed, while achieving a greenhouse environment within defined permissible ranges of air temperature and relative humidity.
KW - Dynamic neural network model
KW - Evaporative fog cooling
KW - Greenhouse climate control
UR - http://www.scopus.com/inward/record.url?scp=84863667291&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84863667291&partnerID=8YFLogxK
U2 - 10.17660/ActaHortic.2012.952.2
DO - 10.17660/ActaHortic.2012.952.2
M3 - Conference contribution
AN - SCOPUS:84863667291
SN - 9789066053380
T3 - Acta Horticulturae
SP - 45
EP - 52
BT - International Symposium on Advanced Technologies and Management Towards Sustainable Greenhouse Ecosystems
PB - International Society for Horticultural Science
ER -